Auxiliary function as user interface for fine-tuning RDS.bootstrap.intervals algorithm, which computes interval estimates for via bootstrapping.
control.rds.estimates(
confidence.level = 0.95,
SS.infinity = 0.01,
lowprevalence = c(8, 14),
discrete.cutoff = 0.8,
useC = TRUE,
number.of.bootstrap.samples = NULL,
hcg.reltol = sqrt(.Machine$double.eps),
hcg.BS.reltol = 1e+05 * sqrt(.Machine$double.eps),
hcg.max.optim = 500,
seed = NULL
)A list with arguments as components.
The confidence level for the confidence intervals. The default is 0.95 for 95%.
The sample proportion, n/N, below which the computation
of the SS weights should simplify to that of the RDS-II weights.
Standard confidence interval procedures can be inaccurate when the
outcome expected count is close to zero. This sets conditions where alternatives to the
standard are used for the ci.type="hmg" option. See Details for its use.
The minimum proportion of the values of the outcome variable that need to be unique before the variable is judged to be continuous.
Use a C-level implementation of Gile's bootstrap (rather than the R level). The implementations should be computational equivalent (except for speed).
The number of bootstrap samples to take
in estimating the uncertainty of the estimator. If NULL it defaults
to the number necessary to compute the standard error to accuracy 0.001.
Relative convergence tolerance for the HCG estimator. The algorithm stops if
it is unable to reduce the log-likelihood by a factor of reltol * (abs(log-likelihood) + reltol)
at a step. Defaults to sqrt(.Machine$double.eps), typically about 1e-8.
Relative convergence tolerance for the bootstrap of the HCG estimator.
It has the same interpretation as hcg.reltol except it is applied to each bootstrap sample.
It is typically the same or larger than hcg.reltol.
The number of iterations on the likelihood optimization for the HCG estimator.
Seed value (integer) for the random number generator. See
set.seed
This function is only used within a call to the RDS.bootstrap.intervals
function.
Some of the arguments are not yet fully implemented. It will evolve slower to incorporate more arguments as the package develops.
Standard confidence interval procedures can be inaccurate when the
outcome expected count is close to zero. In these cases
the combined Agresti-Coull and the bootstrap-t interval of
Mantalos and Zografos (2008) can be used.
The lowprevalence argument is a
two vector parameter setting the conditions under which the approximation is used.
The first is the penalty term on the differential activity. If the observed number
of the rare group minus the product of the first parameter and the differential
activity is lower than the
second parameter, the low prevalence approximation is used.
RDS.bootstrap.intervals